Skip to contentSkip to conversation

iommu/vt-d: Improve PASID id and table management

PATCH Superseded Lu Baolu 3 participants 16 messages May 4, 2018 – May 30, 2018 IOMMU

v2 of 5 Superseded Latest is v5 →

lore
Share & tools
lore

Lu Baolu

author
#0 lore raw

v2 cover iommu/vt-d: Improve PASID id and table management

This patch set improves the PASID id and PASID table management for Intel IOMMU driver.

PATCH 1~3 replace per IOMMU idr name space with a global one. Current per IOMMU idr doesn't work in some cases where one application (associated with a PASID) might talk to two physical devices simultaneously while the two devices could reside behind two different IOMMU units.

PATCH 4~9 implement per domain PASID table. Current per IOMMU PASID table implementation is insecure in the cases where multiple devices under one single IOMMU unit support PASID feature. With per domain PASID table, we can achieve finer protection and isolation granularity.

Best regards,
Lu Baolu

Change log:
v1->v2:
- Patches have been reviewed by "Liu Yi L <[email protected]>".
- An error case handling was added in PATCH 6/9.
- Some commit messages are refined to be more accurate.

Lu Baolu (9):
iommu/vt-d: Global PASID name space iommu/vt-d: Decouple idr bond pointer from svm iommu/vt-d: Use global PASID for SVM usage iommu/vt-d: Move device_domain_info to header iommu/vt-d: Per domain pasid table interfaces iommu/vt-d: Allocate and free pasid table iommu/vt-d: Calculate PTS value
iommu/vt-d: Use per-domain pasid table iommu/vt-d: Clean up PASID talbe management for SVM

 drivers/iommu/Makefile      |   2 +-
 drivers/iommu/intel-iommu.c | 128 ++++++++++++++++-------------------------
 drivers/iommu/intel-pasid.c | 135 ++++++++++++++++++++++++++++++++++++++++++++
 drivers/iommu/intel-pasid.h |  34 +++++++++++
 drivers/iommu/intel-svm.c   |  98 ++++++++++++++++----------------
 include/linux/intel-iommu.h |  90 +++++++++++++++++++++++++++--
 6 files changed, 352 insertions(+), 135 deletions(-)
 create mode 100644 drivers/iommu/intel-pasid.c
 create mode 100644 drivers/iommu/intel-pasid.h
Signature · 1 line
2.7.4
author
#1 lore raw

PATCH v2 1/9 iommu/vt-d: Global PASID name space

This adds the system wide PASID name space for the PASID allocation. Currently we are using per IOMMU PASID name spaces which are not suitable for some use cases. For an example, one application (associated with a PASID) might talk to two physical devices simultaneously while the two devices could reside behind two different IOMMU units.

---
 drivers/iommu/Makefile      |  2 +-
 drivers/iommu/intel-iommu.c | 13 ++++++++++
 drivers/iommu/intel-pasid.c | 60 +++++++++++++++++++++++++++++++++++++++++++++
 drivers/iommu/intel-pasid.h | 30 +++++++++++++++++++++++
 4 files changed, 104 insertions(+), 1 deletion(-)
 create mode 100644 drivers/iommu/intel-pasid.c
 create mode 100644 drivers/iommu/intel-pasid.h
drivers/iommu/Makefile +1 -1
index 1fb6958..0a190b4 100644
@@ -14,7 +14,7 @@ obj-$(CONFIG_AMD_IOMMU_V2) += amd_iommu_v2.o
 obj-$(CONFIG_ARM_SMMU) += arm-smmu.o
 obj-$(CONFIG_ARM_SMMU_V3) += arm-smmu-v3.o
 obj-$(CONFIG_DMAR_TABLE) += dmar.o
-obj-$(CONFIG_INTEL_IOMMU) += intel-iommu.o
+obj-$(CONFIG_INTEL_IOMMU) += intel-iommu.o intel-pasid.o
 obj-$(CONFIG_INTEL_IOMMU_SVM) += intel-svm.o
 obj-$(CONFIG_IPMMU_VMSA) += ipmmu-vmsa.o
 obj-$(CONFIG_IRQ_REMAP) += intel_irq_remapping.o irq_remapping.o
drivers/iommu/intel-iommu.c +13 -0
index 749d8f2..98c5ae9 100644
@@ -53,6 +53,7 @@
 #include <asm/iommu.h>
 
 #include "irq_remapping.h"
+#include "intel-pasid.h"
 
 #define ROOT_SIZE		VTD_PAGE_SIZE
 #define CONTEXT_SIZE		VTD_PAGE_SIZE
@@ -3265,6 +3266,18 @@ static int __init init_dmars(void)
 	}
 
 	for_each_active_iommu(iommu, drhd) {
+		/*
+		 * Find the max pasid size of all IOMMU's in the system.
+		 * we need to ensure the system pasid table is no bigger
+		 * than the smallest supported.
+		 */
+		if (pasid_enabled(iommu)) {
+			u32 temp = 2 << ecap_pss(iommu->ecap);
+
+			intel_pasid_max_id = min_t(u32, temp,
+						   intel_pasid_max_id);
+		}
+
 		g_iommus[iommu->seq_id] = iommu;
 
 		intel_iommu_init_qi(iommu);
drivers/iommu/intel-pasid.c new file +60 -0

60 lines not on this page · raw message

drivers/iommu/intel-pasid.h new file +30 -0
new file mode 100644
index 0000000..0c36af0
@@ -0,0 +1,30 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * intel-pasid.h - PASID idr, table and entry header
+ *
+ * Copyright (C) 2018 Intel Corporation
+ *
+ * Author: Lu Baolu <baolu.lu-VuQAYsv1563Yd54FQh9/[email protected]>
+ */
+
+#ifndef __INTEL_PASID_H
+#define __INTEL_PASID_H
+
+/*
+ * Eventually I'm promised we will get a multi-level PASID table
+ * and it won't have to be physically contiguous. Until then,
+ * limit the size because 8MiB contiguous allocations can be hard
+ * to come by. The limit of 0x20000, which is 1MiB for each of
+ * the PASID and PASID-state tables, is somewhat arbitrary.
+ *
+ * PASID 0 is reserved in caching mode (virtualised IOMMU).
+ */
+#define PASID_MIN			0x1
+#define PASID_MAX			0x20000
+
+extern u32 intel_pasid_max_id;
+int intel_pasid_alloc_id(void *ptr, int start, int end, gfp_t gfp);
+void intel_pasid_free_id(int pasid);
+void *intel_pasid_lookup_id(int pasid);
+
+#endif /* __INTEL_PASID_H */
Signature · 1 line
2.7.4
author
#2 lore raw

PATCH v2 2/9 iommu/vt-d: Decouple idr bond pointer from svm

In reply to Lu Baolu

As we move the PASID idr out of SVM code and make it serving as a global PASID name space, the consumer can specify a ptr to bind it with a PASID. We shouldn't assume that each PASID will be bond with a ptr of struct intel_svm anymore.

This patch cleans up a idr_for_each_entry() usage in the SVM code. It's required to replace the SVM-specific idr with the global PASID idr.

Cc Ashok Raj <[email protected]>
Cc Jacob Pan <jacob.jun.pan-VuQAYsv1563Yd54FQh9/[email protected]>
Cc Kevin Tian <[email protected]>
Cc Liu Yi L <[email protected]>
---
 drivers/iommu/intel-svm.c   | 14 ++++++++++----
 include/linux/intel-iommu.h |  1 +
 2 files changed, 11 insertions(+), 4 deletions(-)
drivers/iommu/intel-svm.c +10 -4
index e8cd984..983af0c 100644
@@ -298,6 +298,7 @@ static const struct mmu_notifier_ops intel_mmuops = {
 };
 
 static DEFINE_MUTEX(pasid_mutex);
+static LIST_HEAD(global_svm_list);
 
 int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_ops *ops)
 {
@@ -329,13 +330,13 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 
 	mutex_lock(&pasid_mutex);
 	if (pasid && !(flags & SVM_FLAG_PRIVATE_PASID)) {
-		int i;
+		struct intel_svm *t;
 
-		idr_for_each_entry(&iommu->pasid_idr, svm, i) {
-			if (svm->mm != mm ||
-			    (svm->flags & SVM_FLAG_PRIVATE_PASID))
+		list_for_each_entry(t, &global_svm_list, list) {
+			if (t->mm != mm || (t->flags & SVM_FLAG_PRIVATE_PASID))
 				continue;
 
+			svm = t;
 			if (svm->pasid >= pasid_max) {
 				dev_warn(dev,
 					 "Limited PASID width. Cannot use existing PASID %d\n",
@@ -404,6 +405,7 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		svm->mm = mm;
 		svm->flags = flags;
 		INIT_LIST_HEAD_RCU(&svm->devs);
+		INIT_LIST_HEAD(&svm->list);
 		ret = -ENOMEM;
 		if (mm) {
 			ret = mmu_notifier_register(&svm->notifier, mm);
@@ -430,6 +432,8 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		 */
 		if (cap_caching_mode(iommu->cap))
 			intel_flush_pasid_dev(svm, sdev, svm->pasid);
+
+		list_add_tail(&svm->list, &global_svm_list);
 	}
 	list_add_rcu(&sdev->list, &svm->devs);
 
@@ -485,6 +489,8 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 					if (svm->mm)
 						mmu_notifier_unregister(&svm->notifier, svm->mm);
 
+					list_del(&svm->list);
+
 					/* We mandate that no page faults may be outstanding
 					 * for the PASID when intel_svm_unbind_mm() is called.
 					 * If that is not obeyed, subtle errors will happen.
include/linux/intel-iommu.h +1 -0
index ef169d6..795717e 100644
@@ -486,6 +486,7 @@ struct intel_svm {
 	int flags;
 	int pasid;
 	struct list_head devs;
+	struct list_head list;
 };
 
 extern int intel_iommu_enable_pasid(struct intel_iommu *iommu, struct intel_svm_dev *sdev);
Signature · 1 line
2.7.4
author
#3 lore raw

PATCH v2 3/9 iommu/vt-d: Use global PASID for SVM usage

In reply to Lu Baolu

This patch switches PASID management for SVM from per iommu idr to the global idr.

---
 drivers/iommu/intel-svm.c   | 22 +++++++++++-----------
 include/linux/intel-iommu.h |  1 -
 2 files changed, 11 insertions(+), 12 deletions(-)
drivers/iommu/intel-svm.c +11 -11
index 983af0c..24d0ea1 100644
@@ -26,6 +26,8 @@
 #include <linux/interrupt.h>
 #include <asm/page.h>
 
+#include "intel-pasid.h"
+
 #define PASID_ENTRY_P		BIT_ULL(0)
 #define PASID_ENTRY_FLPM_5LP	BIT_ULL(9)
 #define PASID_ENTRY_SRE		BIT_ULL(11)
@@ -85,8 +87,6 @@ int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
 				iommu->name);
 	}
 
-	idr_init(&iommu->pasid_idr);
-
 	return 0;
 }
 
@@ -102,7 +102,7 @@ int intel_svm_free_pasid_tables(struct intel_iommu *iommu)
 		free_pages((unsigned long)iommu->pasid_state_table, order);
 		iommu->pasid_state_table = NULL;
 	}
-	idr_destroy(&iommu->pasid_idr);
+
 	return 0;
 }
 
@@ -392,9 +392,9 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 			pasid_max = iommu->pasid_max;
 
 		/* Do not use PASID 0 in caching mode (virtualised IOMMU) */
-		ret = idr_alloc(&iommu->pasid_idr, svm,
-				!!cap_caching_mode(iommu->cap),
-				pasid_max - 1, GFP_KERNEL);
+		ret = intel_pasid_alloc_id(svm,
+					   !!cap_caching_mode(iommu->cap),
+					   pasid_max - 1, GFP_KERNEL);
 		if (ret < 0) {
 			kfree(svm);
 			kfree(sdev);
@@ -410,7 +410,7 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		if (mm) {
 			ret = mmu_notifier_register(&svm->notifier, mm);
 			if (ret) {
-				idr_remove(&svm->iommu->pasid_idr, svm->pasid);
+				intel_pasid_free_id(svm->pasid);
 				kfree(svm);
 				kfree(sdev);
 				goto out;
@@ -460,7 +460,7 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 	if (!iommu || !iommu->pasid_table)
 		goto out;
 
-	svm = idr_find(&iommu->pasid_idr, pasid);
+	svm = intel_pasid_lookup_id(pasid);
 	if (!svm)
 		goto out;
 
@@ -485,7 +485,7 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 					svm->iommu->pasid_table[svm->pasid].val = 0;
 					wmb();
 
-					idr_remove(&svm->iommu->pasid_idr, svm->pasid);
+					intel_pasid_free_id(svm->pasid);
 					if (svm->mm)
 						mmu_notifier_unregister(&svm->notifier, svm->mm);
 
@@ -520,7 +520,7 @@ int intel_svm_is_pasid_valid(struct device *dev, int pasid)
 	if (!iommu || !iommu->pasid_table)
 		goto out;
 
-	svm = idr_find(&iommu->pasid_idr, pasid);
+	svm = intel_pasid_lookup_id(pasid);
 	if (!svm)
 		goto out;
 
@@ -618,7 +618,7 @@ static irqreturn_t prq_event_thread(int irq, void *d)
 
 		if (!svm || svm->pasid != req->pasid) {
 			rcu_read_lock();
-			svm = idr_find(&iommu->pasid_idr, req->pasid);
+			svm = intel_pasid_lookup_id(req->pasid);
 			/* It *can't* go away, because the driver is not permitted
 			 * to unbind the mm while any page faults are outstanding.
 			 * So we only need RCU to protect the internal idr code. */
include/linux/intel-iommu.h +0 -1
index 795717e..6b5ef6c 100644
@@ -418,7 +418,6 @@ struct intel_iommu {
 	struct pasid_state_entry *pasid_state_table;
 	struct page_req_dsc *prq;
 	unsigned char prq_name[16];    /* Name for PRQ interrupt */
-	struct idr pasid_idr;
 	u32 pasid_max;
 #endif
 	struct q_inval  *qi;            /* Queued invalidation info */
Signature · 1 line
2.7.4
author
#4 lore raw

PATCH v2 4/9 iommu/vt-d: Move device_domain_info to header

In reply to Lu Baolu

This allows the per device iommu data to be accessed from other files.

Cc Ashok Raj <[email protected]>
Cc Jacob Pan <jacob.jun.pan-VuQAYsv1563Yd54FQh9/[email protected]>
Cc Kevin Tian <[email protected]>
Cc Liu Yi L <[email protected]>
---
 drivers/iommu/intel-iommu.c | 62 +++--------------------------------------
 include/linux/intel-iommu.h | 68 +++++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 72 insertions(+), 58 deletions(-)
drivers/iommu/intel-iommu.c +4 -58
index 98c5ae9..caa0b5c 100644
@@ -381,60 +381,6 @@ static int hw_pass_through = 1;
 	for (idx = 0; idx < g_num_of_iommus; idx++)		\
 		if (domain->iommu_refcnt[idx])
 
-struct dmar_domain {
-	int	nid;			/* node id */
-
-	unsigned	iommu_refcnt[DMAR_UNITS_SUPPORTED];
-					/* Refcount of devices per iommu */
-
-
-	u16		iommu_did[DMAR_UNITS_SUPPORTED];
-					/* Domain ids per IOMMU. Use u16 since
-					 * domain ids are 16 bit wide according
-					 * to VT-d spec, section 9.3 */
-
-	bool has_iotlb_device;
-	struct list_head devices;	/* all devices' list */
-	struct iova_domain iovad;	/* iova's that belong to this domain */
-
-	struct dma_pte	*pgd;		/* virtual address */
-	int		gaw;		/* max guest address width */
-
-	/* adjusted guest address width, 0 is level 2 30-bit */
-	int		agaw;
-
-	int		flags;		/* flags to find out type of domain */
-
-	int		iommu_coherency;/* indicate coherency of iommu access */
-	int		iommu_snooping; /* indicate snooping control feature*/
-	int		iommu_count;	/* reference count of iommu */
-	int		iommu_superpage;/* Level of superpages supported:
-					   0 == 4KiB (no superpages), 1 == 2MiB,
-					   2 == 1GiB, 3 == 512GiB, 4 == 1TiB */
-	u64		max_addr;	/* maximum mapped address */
-
-	struct iommu_domain domain;	/* generic domain data structure for
-					   iommu core */
-};
-
-/* PCI domain-device relationship */
-struct device_domain_info {
-	struct list_head link;	/* link to domain siblings */
-	struct list_head global; /* link to global list */
-	u8 bus;			/* PCI bus number */
-	u8 devfn;		/* PCI devfn number */
-	u8 pasid_supported:3;
-	u8 pasid_enabled:1;
-	u8 pri_supported:1;
-	u8 pri_enabled:1;
-	u8 ats_supported:1;
-	u8 ats_enabled:1;
-	u8 ats_qdep;
-	struct device *dev; /* it's NULL for PCIe-to-PCI bridge */
-	struct intel_iommu *iommu; /* IOMMU used by this device */
-	struct dmar_domain *domain; /* pointer to domain */
-};
-
 struct dmar_rmrr_unit {
 	struct list_head list;		/* list of rmrr units	*/
 	struct acpi_dmar_header *hdr;	/* ACPI header		*/
@@ -631,7 +577,7 @@ static void set_iommu_domain(struct intel_iommu *iommu, u16 did,
 		domains[did & 0xff] = domain;
 }
 
-static inline void *alloc_pgtable_page(int node)
+void *alloc_pgtable_page(int node)
 {
 	struct page *page;
 	void *vaddr = NULL;
@@ -642,7 +588,7 @@ static inline void *alloc_pgtable_page(int node)
 	return vaddr;
 }
 
-static inline void free_pgtable_page(void *vaddr)
+void free_pgtable_page(void *vaddr)
 {
 	free_page((unsigned long)vaddr);
 }
@@ -725,7 +671,7 @@ int iommu_calculate_agaw(struct intel_iommu *iommu)
 }
 
 /* This functionin only returns single iommu in a domain */
-static struct intel_iommu *domain_get_iommu(struct dmar_domain *domain)
+struct intel_iommu *domain_get_iommu(struct dmar_domain *domain)
 {
 	int iommu_id;
 
@@ -3500,7 +3446,7 @@ static unsigned long intel_alloc_iova(struct device *dev,
 	return iova_pfn;
 }
 
-static struct dmar_domain *get_valid_domain_for_dev(struct device *dev)
+struct dmar_domain *get_valid_domain_for_dev(struct device *dev)
 {
 	struct dmar_domain *domain, *tmp;
 	struct dmar_rmrr_unit *rmrr;
include/linux/intel-iommu.h +68 -0

92 lines not on this page · raw message

Signature · 1 line
2.7.4
author
#5 lore raw

PATCH v2 5/9 iommu/vt-d: Per domain pasid table interfaces

In reply to Lu Baolu

This patch adds the interfaces for per domain pasid table management. Currently we allocate one pasid table for all devices under the scope of an IOMMU. It's insecure in the cases where multiple devices under one single IOMMU unit support PASID feature. With per domain pasid table, we can achieve finer protection and isolation granularity.

Cc Ashok Raj <[email protected]>
Cc Jacob Pan <jacob.jun.pan-VuQAYsv1563Yd54FQh9/[email protected]>
Cc Kevin Tian <[email protected]>
Cc Liu Yi L <[email protected]>
---
 drivers/iommu/intel-pasid.c | 75 +++++++++++++++++++++++++++++++++++++++++++++
 drivers/iommu/intel-pasid.h |  4 +++
 include/linux/intel-iommu.h |  5 +++
 3 files changed, 84 insertions(+)
drivers/iommu/intel-pasid.c +75 -0
index 0690f39..b8691a6 100644
@@ -13,6 +13,7 @@
 #include <linux/intel-iommu.h>
 #include <linux/iommu.h>
 #include <linux/memory.h>
+#include <linux/pci.h>
 #include <linux/spinlock.h>
 
 #include "intel-pasid.h"
@@ -58,3 +59,77 @@ void *intel_pasid_lookup_id(int pasid)
 
 	return p;
 }
+
+/*
+ * Interfaces for per domain pasid table management:
+ */
+int intel_pasid_alloc_table(struct device *dev, size_t entry_size,
+			    size_t entry_count)
+{
+	struct device_domain_info *info;
+	struct dmar_domain *domain;
+	struct page *pages;
+	int order;
+
+	info = dev->archdata.iommu;
+	if (WARN_ON(!info || !dev_is_pci(dev) ||
+		    !info->pasid_supported ||
+		    !info->domain))
+		return -EINVAL;
+
+	domain = info->domain;
+
+	if (entry_count > intel_pasid_max_id)
+		entry_count = intel_pasid_max_id;
+
+	order = get_order(entry_size * entry_count);
+	pages = alloc_pages_node(domain->nid, GFP_KERNEL | __GFP_ZERO, order);
+	if (!pages)
+		return -ENOMEM;
+
+	spin_lock(&pasid_lock);
+	if (domain->pasid_table) {
+		__free_pages(pages, order);
+	} else {
+		domain->pasid_table	= page_address(pages);
+		domain->order		= order;
+		domain->max_pasid	= entry_count;
+	}
+	domain->pasid_users++;
+	spin_unlock(&pasid_lock);
+
+	return 0;
+}
+
+void intel_pasid_free_table(struct device *dev)
+{
+	struct dmar_domain *domain;
+
+	domain = get_valid_domain_for_dev(dev);
+	if (!domain || !dev_is_pci(dev))
+		return;
+
+	spin_lock(&pasid_lock);
+	if (domain->pasid_table) {
+		domain->pasid_users--;
+		if (!domain->pasid_users) {
+			free_pages((unsigned long)domain->pasid_table,
+				   domain->order);
+			domain->pasid_table	= NULL;
+			domain->order		= 0;
+			domain->max_pasid	= 0;
+		}
+	}
+	spin_unlock(&pasid_lock);
+}
+
+void *intel_pasid_get_table(struct device *dev)
+{
+	struct dmar_domain *domain;
+
+	domain = get_valid_domain_for_dev(dev);
+	if (!domain)
+		return NULL;
+
+	return domain->pasid_table;
+}
drivers/iommu/intel-pasid.h +4 -0
index 0c36af0..a90c60b 100644
@@ -26,5 +26,9 @@ extern u32 intel_pasid_max_id;
 int intel_pasid_alloc_id(void *ptr, int start, int end, gfp_t gfp);
 void intel_pasid_free_id(int pasid);
 void *intel_pasid_lookup_id(int pasid);
+int intel_pasid_alloc_table(struct device *dev, size_t entry_size,
+			    size_t entry_count);
+void intel_pasid_free_table(struct device *dev);
+void *intel_pasid_get_table(struct device *dev);
 
 #endif /* __INTEL_PASID_H */
include/linux/intel-iommu.h +5 -0
index a4463f0..bee7a3f 100644
@@ -424,6 +424,11 @@ struct dmar_domain {
 					 */
 	u64		max_addr;	/* maximum mapped address */
 
+	void		*pasid_table;	/* pointer of pasid table */
+	unsigned int	pasid_users;	/* User number of pasid table */
+	int		order;		/* the page order of tables */
+	int		max_pasid;	/* max pasid */
+
 	struct iommu_domain domain;	/*
 					 * generic domain data structure for
 					 * iommu core
Signature · 1 line
2.7.4
author
#6 lore raw

PATCH v2 6/9 iommu/vt-d: Allocate and free pasid table

In reply to Lu Baolu

This patch allocates PASID table for a domain at the time when it is being created (if any devices using this domain supports PASID feature), and free it when the domain is freed.

---
 drivers/iommu/intel-iommu.c | 19 +++++++++++++++++++
 drivers/iommu/intel-svm.c   |  8 --------
 include/linux/intel-iommu.h | 10 ++++++++--
 3 files changed, 27 insertions(+), 10 deletions(-)
drivers/iommu/intel-iommu.c +19 -0
index caa0b5c..f86302d 100644
@@ -2460,6 +2460,24 @@ static struct dmar_domain *dmar_insert_one_dev_info(struct intel_iommu *iommu,
 		dev->archdata.iommu = info;
 	spin_unlock_irqrestore(&device_domain_lock, flags);
 
+	if (dev && dev_is_pci(dev) && info->pasid_supported) {
+		if (pasid_enabled(iommu)) {
+			size_t size, count;
+
+			size = sizeof(struct pasid_entry);
+			count = min_t(int,
+				      pci_max_pasids(to_pci_dev(dev)),
+				      intel_pasid_max_id);
+			ret = intel_pasid_alloc_table(dev, size, count);
+			if (ret) {
+				pr_err("PASID table allocation for %s failed\n",
+				       dev_name(dev));
+				dmar_remove_one_dev_info(domain, dev);
+				return NULL;
+			}
+		}
+	}
+
 	if (dev && domain_context_mapping(domain, dev)) {
 		pr_err("Domain context map for %s failed\n", dev_name(dev));
 		dmar_remove_one_dev_info(domain, dev);
@@ -4826,6 +4844,7 @@ static void dmar_remove_one_dev_info(struct dmar_domain *domain,
 	unsigned long flags;
 
 	spin_lock_irqsave(&device_domain_lock, flags);
+	intel_pasid_free_table(dev);
 	info = dev->archdata.iommu;
 	__dmar_remove_one_dev_info(info);
 	spin_unlock_irqrestore(&device_domain_lock, flags);
drivers/iommu/intel-svm.c +0 -8
index 24d0ea1..3abc94f 100644
@@ -34,14 +34,6 @@
 
 static irqreturn_t prq_event_thread(int irq, void *d);
 
-struct pasid_entry {
-	u64 val;
-};
-
-struct pasid_state_entry {
-	u64 val;
-};
-
 int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
 {
 	struct page *pages;
include/linux/intel-iommu.h +8 -2
index bee7a3f..08e5811 100644
@@ -382,8 +382,14 @@ enum {
 #define VTD_FLAG_TRANS_PRE_ENABLED	(1 << 0)
 #define VTD_FLAG_IRQ_REMAP_PRE_ENABLED	(1 << 1)
 
-struct pasid_entry;
-struct pasid_state_entry;
+struct pasid_entry {
+	u64 val;
+};
+
+struct pasid_state_entry {
+	u64 val;
+};
+
 struct page_req_dsc;
 
 struct dmar_domain {
Signature · 1 line
2.7.4
author
#7 lore raw

PATCH v2 7/9 iommu/vt-d: Calculate PTS value

In reply to Lu Baolu

Calculate PTS (PASID Table Size) value for the extended context entry from the real size of the PASID table for a domain.

Cc Ashok Raj <[email protected]>
Cc Jacob Pan <jacob.jun.pan-VuQAYsv1563Yd54FQh9/[email protected]>
Cc Kevin Tian <[email protected]>
Cc Liu Yi L <[email protected]>
---
 drivers/iommu/intel-iommu.c | 22 ++++++++--------------
 1 file changed, 8 insertions(+), 14 deletions(-)
drivers/iommu/intel-iommu.c +8 -14
index f86302d..5602ccd 100644
@@ -5152,22 +5152,16 @@ static void intel_iommu_put_resv_regions(struct device *dev,
 
 #ifdef CONFIG_INTEL_IOMMU_SVM
 #define MAX_NR_PASID_BITS (20)
-static inline unsigned long intel_iommu_get_pts(struct intel_iommu *iommu)
+static inline unsigned long intel_iommu_get_pts(struct dmar_domain *domain)
 {
-	/*
-	 * Convert ecap_pss to extend context entry pts encoding, also
-	 * respect the soft pasid_max value set by the iommu.
-	 * - number of PASID bits = ecap_pss + 1
-	 * - number of PASID table entries = 2^(pts + 5)
-	 * Therefore, pts = ecap_pss - 4
-	 * e.g. KBL ecap_pss = 0x13, PASID has 20 bits, pts = 15
-	 */
-	if (ecap_pss(iommu->ecap) < 5)
+	int pts;
+
+	pts = find_first_bit((unsigned long *)&domain->max_pasid,
+			     MAX_NR_PASID_BITS);
+	if (pts < 5)
 		return 0;
 
-	/* pasid_max is encoded as actual number of entries not the bits */
-	return find_first_bit((unsigned long *)&iommu->pasid_max,
-			MAX_NR_PASID_BITS) - 5;
+	return pts - 5;
 }
 
 int intel_iommu_enable_pasid(struct intel_iommu *iommu, struct intel_svm_dev *sdev)
@@ -5204,7 +5198,7 @@ int intel_iommu_enable_pasid(struct intel_iommu *iommu, struct intel_svm_dev *sd
 		if (iommu->pasid_state_table)
 			context[1].hi = (u64)virt_to_phys(iommu->pasid_state_table);
 		context[1].lo = (u64)virt_to_phys(iommu->pasid_table) |
-			intel_iommu_get_pts(iommu);
+			intel_iommu_get_pts(domain);
 
 		wmb();
 		/* CONTEXT_TT_MULTI_LEVEL and CONTEXT_TT_DEV_IOTLB are both
Signature · 1 line
2.7.4
author
#8 lore raw

PATCH v2 8/9 iommu/vt-d: Use per-domain pasid table

In reply to Lu Baolu

This patch replaces current per iommu pasid table with the new added per domain pasid table. Each svm-capable PCI device will be configured with a pasid table which shares with other svm-capable devices within its iommu domain.

---
 drivers/iommu/intel-iommu.c |  6 +++---
 drivers/iommu/intel-svm.c   | 37 +++++++++++++++++++++++++------------
 2 files changed, 28 insertions(+), 15 deletions(-)
drivers/iommu/intel-iommu.c +3 -3
index 5602ccd..fa6052a 100644
@@ -5197,7 +5197,7 @@ int intel_iommu_enable_pasid(struct intel_iommu *iommu, struct intel_svm_dev *sd
 	if (!(ctx_lo & CONTEXT_PASIDE)) {
 		if (iommu->pasid_state_table)
 			context[1].hi = (u64)virt_to_phys(iommu->pasid_state_table);
-		context[1].lo = (u64)virt_to_phys(iommu->pasid_table) |
+		context[1].lo = (u64)virt_to_phys(domain->pasid_table) |
 			intel_iommu_get_pts(domain);
 
 		wmb();
@@ -5265,8 +5265,8 @@ struct intel_iommu *intel_svm_device_to_iommu(struct device *dev)
 		return NULL;
 	}
 
-	if (!iommu->pasid_table) {
-		dev_err(dev, "PASID not enabled on IOMMU; cannot enable SVM\n");
+	if (!intel_pasid_get_table(dev)) {
+		dev_err(dev, "No PASID table for device; cannot enable SVM\n");
 		return NULL;
 	}
 
drivers/iommu/intel-svm.c +25 -12
index 3abc94f..3b14819 100644
@@ -256,6 +256,7 @@ static void intel_flush_pasid_dev(struct intel_svm *svm, struct intel_svm_dev *s
 static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
 {
 	struct intel_svm *svm = container_of(mn, struct intel_svm, notifier);
+	struct pasid_entry *pasid_table;
 	struct intel_svm_dev *sdev;
 
 	/* This might end up being called from exit_mmap(), *before* the page
@@ -270,11 +271,16 @@ static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
 	 * page) so that we end up taking a fault that the hardware really
 	 * *has* to handle gracefully without affecting other processes.
 	 */
-	svm->iommu->pasid_table[svm->pasid].val = 0;
-	wmb();
-
 	rcu_read_lock();
 	list_for_each_entry_rcu(sdev, &svm->devs, list) {
+		pasid_table = intel_pasid_get_table(sdev->dev);
+		if (!pasid_table)
+			continue;
+
+		pasid_table[svm->pasid].val = 0;
+		/* Make sure the entry update is visible before translation. */
+		wmb();
+
 		intel_flush_pasid_dev(svm, sdev, svm->pasid);
 		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0, !svm->mm);
 	}
@@ -295,6 +301,7 @@ static LIST_HEAD(global_svm_list);
 int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_ops *ops)
 {
 	struct intel_iommu *iommu = intel_svm_device_to_iommu(dev);
+	struct pasid_entry *pasid_table;
 	struct intel_svm_dev *sdev;
 	struct intel_svm *svm = NULL;
 	struct mm_struct *mm = NULL;
@@ -302,7 +309,8 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 	int pasid_max;
 	int ret;
 
-	if (WARN_ON(!iommu || !iommu->pasid_table))
+	pasid_table = intel_pasid_get_table(dev);
+	if (WARN_ON(!iommu || !pasid_table))
 		return -EINVAL;
 
 	if (dev_is_pci(dev)) {
@@ -380,8 +388,8 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		}
 		svm->iommu = iommu;
 
-		if (pasid_max > iommu->pasid_max)
-			pasid_max = iommu->pasid_max;
+		if (pasid_max > intel_pasid_max_id)
+			pasid_max = intel_pasid_max_id;
 
 		/* Do not use PASID 0 in caching mode (virtualised IOMMU) */
 		ret = intel_pasid_alloc_id(svm,
@@ -414,7 +422,7 @@ int intel_svm_bind_mm(struct device *dev, int *pasid, int flags, struct svm_dev_
 		if (cpu_feature_enabled(X86_FEATURE_LA57))
 			pasid_entry_val |= PASID_ENTRY_FLPM_5LP;
 
-		iommu->pasid_table[svm->pasid].val = pasid_entry_val;
+		pasid_table[svm->pasid].val = pasid_entry_val;
 
 		wmb();
 
@@ -442,6 +450,7 @@ EXPORT_SYMBOL_GPL(intel_svm_bind_mm);
 
 int intel_svm_unbind_mm(struct device *dev, int pasid)
 {
+	struct pasid_entry *pasid_table;
 	struct intel_svm_dev *sdev;
 	struct intel_iommu *iommu;
 	struct intel_svm *svm;
@@ -449,7 +458,8 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 
 	mutex_lock(&pasid_mutex);
 	iommu = intel_svm_device_to_iommu(dev);
-	if (!iommu || !iommu->pasid_table)
+	pasid_table = intel_pasid_get_table(dev);
+	if (!iommu || !pasid_table)
 		goto out;
 
 	svm = intel_pasid_lookup_id(pasid);
@@ -472,11 +482,14 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 				intel_flush_pasid_dev(svm, sdev, svm->pasid);
 				intel_flush_svm_range_dev(svm, sdev, 0, -1, 0, !svm->mm);
 				kfree_rcu(sdev, rcu);
+				pasid_table[svm->pasid].val = 0;
+				/*
+				 * Make sure the entry update is visible
+				 * before translation.
+				 */
+				wmb();
 
 				if (list_empty(&svm->devs)) {
-					svm->iommu->pasid_table[svm->pasid].val = 0;
-					wmb();
-
 					intel_pasid_free_id(svm->pasid);
 					if (svm->mm)
 						mmu_notifier_unregister(&svm->notifier, svm->mm);
@@ -509,7 +522,7 @@ int intel_svm_is_pasid_valid(struct device *dev, int pasid)
 
 	mutex_lock(&pasid_mutex);
 	iommu = intel_svm_device_to_iommu(dev);
-	if (!iommu || !iommu->pasid_table)
+	if (!iommu)
 		goto out;
 
 	svm = intel_pasid_lookup_id(pasid);
Signature · 1 line
2.7.4
author
#9 lore raw

PATCH v2 9/9 iommu/vt-d: Clean up PASID talbe management for SVM

In reply to Lu Baolu

The previous per iommu pasid table alloc/free interfaces are no longer used. Clean up the driver by removing it.

---
 drivers/iommu/intel-iommu.c |  6 +++---
 drivers/iommu/intel-svm.c   | 17 ++---------------
 include/linux/intel-iommu.h |  5 ++---
 3 files changed, 7 insertions(+), 21 deletions(-)
drivers/iommu/intel-iommu.c +3 -3
index fa6052a..010871d 100644
@@ -1736,7 +1736,7 @@ static void free_dmar_iommu(struct intel_iommu *iommu)
 	if (pasid_enabled(iommu)) {
 		if (ecap_prs(iommu->ecap))
 			intel_svm_finish_prq(iommu);
-		intel_svm_free_pasid_tables(iommu);
+		intel_svm_exit(iommu);
 	}
 #endif
 }
@@ -3297,7 +3297,7 @@ static int __init init_dmars(void)
 			hw_pass_through = 0;
 #ifdef CONFIG_INTEL_IOMMU_SVM
 		if (pasid_enabled(iommu))
-			intel_svm_alloc_pasid_tables(iommu);
+			intel_svm_init(iommu);
 #endif
 	}
 
@@ -4274,7 +4274,7 @@ static int intel_iommu_add(struct dmar_drhd_unit *dmaru)
 
 #ifdef CONFIG_INTEL_IOMMU_SVM
 	if (pasid_enabled(iommu))
-		intel_svm_alloc_pasid_tables(iommu);
+		intel_svm_init(iommu);
 #endif
 
 	if (dmaru->ignored) {
drivers/iommu/intel-svm.c +2 -15
index 3b14819..38cae65 100644
@@ -34,7 +34,7 @@
 
 static irqreturn_t prq_event_thread(int irq, void *d);
 
-int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
+int intel_svm_init(struct intel_iommu *iommu)
 {
 	struct page *pages;
 	int order;
@@ -59,15 +59,6 @@ int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
 		iommu->pasid_max = 0x20000;
 
 	order = get_order(sizeof(struct pasid_entry) * iommu->pasid_max);
-	pages = alloc_pages(GFP_KERNEL | __GFP_ZERO, order);
-	if (!pages) {
-		pr_warn("IOMMU: %s: Failed to allocate PASID table\n",
-			iommu->name);
-		return -ENOMEM;
-	}
-	iommu->pasid_table = page_address(pages);
-	pr_info("%s: Allocated order %d PASID table.\n", iommu->name, order);
-
 	if (ecap_dis(iommu->ecap)) {
 		/* Just making it explicit... */
 		BUILD_BUG_ON(sizeof(struct pasid_entry) != sizeof(struct pasid_state_entry));
@@ -82,14 +73,10 @@ int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu)
 	return 0;
 }
 
-int intel_svm_free_pasid_tables(struct intel_iommu *iommu)
+int intel_svm_exit(struct intel_iommu *iommu)
 {
 	int order = get_order(sizeof(struct pasid_entry) * iommu->pasid_max);
 
-	if (iommu->pasid_table) {
-		free_pages((unsigned long)iommu->pasid_table, order);
-		iommu->pasid_table = NULL;
-	}
 	if (iommu->pasid_state_table) {
 		free_pages((unsigned long)iommu->pasid_state_table, order);
 		iommu->pasid_state_table = NULL;
include/linux/intel-iommu.h +2 -3
index 08e5811..44c7613 100644
@@ -470,7 +470,6 @@ struct intel_iommu {
 	 * devices away to userspace processes (e.g. for DPDK) and don't
 	 * want to trust that userspace will use *only* the PASID it was
 	 * told to. But while it's all driver-arbitrated, we're fine. */
-	struct pasid_entry *pasid_table;
 	struct pasid_state_entry *pasid_state_table;
 	struct page_req_dsc *prq;
 	unsigned char prq_name[16];    /* Name for PRQ interrupt */
@@ -539,8 +538,8 @@ void free_pgtable_page(void *vaddr);
 struct intel_iommu *domain_get_iommu(struct dmar_domain *domain);
 
 #ifdef CONFIG_INTEL_IOMMU_SVM
-extern int intel_svm_alloc_pasid_tables(struct intel_iommu *iommu);
-extern int intel_svm_free_pasid_tables(struct intel_iommu *iommu);
+int intel_svm_init(struct intel_iommu *iommu);
+int intel_svm_exit(struct intel_iommu *iommu);
 extern int intel_svm_enable_prq(struct intel_iommu *iommu);
 extern int intel_svm_finish_prq(struct intel_iommu *iommu);
 
Signature · 1 line
2.7.4
maintainer
#10 lore raw
In reply to Lu Baolu
Quoted 5 lines · Lu Baolu

PATCH 4~9 implement per domain PASID table. Current per IOMMU PASID table implementation is insecure in the cases where multiple devices under one single IOMMU unit support PASID feature. With per domain PASID table, we can achieve finer protection and isolation granularity.

Hold on, we hat discussions in the past about doing a system-wide pasid space, so that every mm_struct with devices attached gets the same pasid across all devices it is talking to. Reason was that some devices (will) require this to work correctly. This goes into the opposite direction, so I am a bit confused here. Please explain, is this not longer necessary?

Regards,

Joerg

author
#11 lore raw

Hi Joerg,

Thank you for looking at my patches.

Quoted 7 lines · Joerg Roedel
Quoted 5 lines · Lu Baolu

PATCH 4~9 implement per domain PASID table. Current per IOMMU PASID table implementation is insecure in the cases where multiple devices under one single IOMMU unit support PASID feature. With per domain PASID table, we can achieve finer protection and isolation granularity.

Quoted 6 lines

Hold on, we hat discussions in the past about doing a system-wide pasid space, so that every mm_struct with devices attached gets the same pasid across all devices it is talking to. Reason was that some devices (will) require this to work correctly. This goes into the opposite direction, so I am a bit confused here. Please explain, is this not longer necessary?

You are right. System-wide pasid space is necessary, hence PATCH 1~3 implement it. But PATCH 4~9 don't go into the opposite direction, it's designed to address another potential issue.

With system-wide pasid space, we can use a system-wide pasid table, or just keep what we have now(per iommu unit pasid table). Both system-wide and per iommu unitpasid table mean that two devices might share a single pasid table. That will result in an issue.

For an example, device A is assigned to access the memory space of process A, and device B is assigned to access the memory space of process B. The dma remapping infrastructure looks like:

                                                     .------------------.
                             .----------------.      |                  |
                             |                |      |                  |
                             .----------------.      | Paging structure |
                             |     PASID X    |--|   |  for Process A   |
                             .----------------.  |   |                  |
                             |                |  --->'------------------'
     .----------------.      .----------------.
     |                |      |     PASID Y    |--|
     .----------------.      .----------------.  |
     | Dev_A context  |---|  |                |  |   .------------------.
     '----------------'   |  .----------------.  |   |                  |
     |                |   |  |                |  |   |                  |
     '----------------'   |  .----------------.  |   | Paging structure |
     | Dev_B context  |   -->|                |  |   |  for Process B   |
     '----------------'----->'----------------'  |   |                  |
     |                |         system-wide      v-->'------------------'
     .----------------.         pasid table    
     |                |                        
     '----------------'
        Intel iommu
       context table

Since dev_A and dev_B share a pasid table, the side effect is that a flawed dev_A might access the memory space of process B (with pasid y). Vice versa, a flawed dev_B might access memory space of process A (with pasid x).

What PATCH 4~9 do is to remove such possibility by assigning a pasid table for each pci device. Hence, the remapping infrastructure looks like:

                                                    .------------------.
                                                    |                  |
                            .----------------.      |                  |
                            |                |      | Paging structure |
                            .----------------.      |  for Process A   |
                            |     PASID X    |      |                  |
                            .----------------.----->'------------------'
                            |                |
                            .----------------.
                            |                |
                            .----------------.
                            |                |
                            .----------------.
   .----------------.       |                |
   |                |       .----------------.
   .----------------.       |                |
   | Dev_A context  |------>'----------------'
   '----------------'          pasid table 
   |                |          for Dev_A   
   '----------------'                      
   | Dev_B context  |-->
   '----------------'  |    .----------------.
   |                |  |    |                |      .------------------.
   .----------------.  |    .----------------.      |                  |
   |                |  |    |                |      |                  |
   '----------------'  |    .----------------.      | Paging structure |
       Intel iommu     |    |                |      |  for Process B   |
      context table    |    .----------------.      |                  |
                       |    |     PASID Y    |----->'------------------'
                       |    .----------------.
                       |    |                |
                       |    .----------------.
                       |    |                |
                       |    .----------------.
                       v--->|                |
                            '----------------'
                               pasid table 
                               for Dev_B   

With this, dev_A has no means to access memory of process B and vice versa.

Best regards,
Lu Baolu

#12 lore raw
Quoted 22 lines

From: Lu Baolu [mailto:baolu.lu-VuQAYsv1563Yd54FQh9/[email protected]]
Sent: Wednesday, May 16, 2018 4:01 PM

Hi Joerg,

Thank you for looking at my patches.

Quoted 7 lines · Joerg Roedel
Quoted 5 lines · Lu Baolu

PATCH 4~9 implement per domain PASID table. Current per IOMMU PASID table implementation is insecure in the cases where multiple devices under one single IOMMU unit support PASID feature. With per domain PASID table, we can achieve finer protection and isolation granularity.

Quoted 6 lines

Hold on, we hat discussions in the past about doing a system-wide pasid space, so that every mm_struct with devices attached gets the same pasid across all devices it is talking to. Reason was that some devices (will) require this to work correctly. This goes into the opposite direction, so I am a bit confused here. Please explain, is this not longer necessary?

Quoted 3 lines

You are right. System-wide pasid space is necessary, hence PATCH 1~3 implement it. But PATCH 4~9 don't go into the opposite direction, it's designed to address another potential issue.

one thing you may want to highlight is, even with PATCH 4-9 it's still doing system-wide PASID space allocation. Just PASID table itself is kept per-device for isolation purpose as you described below, i.e. each device can access only those PASIDs which are allocated to itself while the allocation happens system-wide...

Quoted 88 lines

With system-wide pasid space, we can use a system-wide pasid table, or just keep what we have now(per iommu unit pasid table). Both system-wide and per iommu unitpasid table mean that two devices might share a single pasid table. That will result in an issue.

For an example, device A is assigned to access the memory space of process A, and device B is assigned to access the memory space of process B. The dma remapping infrastructure looks like:

                                                     .------------------.
                             .----------------.      |                  |
                             |                |      |                  |
                             .----------------.      | Paging structure |
                             |     PASID X    |--|   |  for Process A   |
                             .----------------.  |   |                  |
                             |                |  --->'------------------'
     .----------------.      .----------------.
     |                |      |     PASID Y    |--|
     .----------------.      .----------------.  |
     | Dev_A context  |---|  |                |  |   .------------------.
     '----------------'   |  .----------------.  |   |                  |
     |                |   |  |                |  |   |                  |
     '----------------'   |  .----------------.  |   | Paging structure |
     | Dev_B context  |   -->|                |  |   |  for Process B   |
     '----------------'----->'----------------'  |   |                  |
     |                |         system-wide      v-->'------------------'
     .----------------.         pasid table
     |                |
     '----------------'
        Intel iommu
       context table

Since dev_A and dev_B share a pasid table, the side effect is that a flawed dev_A might access the memory space of process B (with pasid y). Vice versa,
a flawed dev_B might access memory space of process A (with pasid x).

What PATCH 4~9 do is to remove such possibility by assigning a pasid table for each pci device. Hence, the remapping infrastructure looks like:

                                                    .------------------.
                                                    |                  |
                            .----------------.      |                  |
                            |                |      | Paging structure |
                            .----------------.      |  for Process A   |
                            |     PASID X    |      |                  |
                            .----------------.----->'------------------'
                            |                |
                            .----------------.
                            |                |
                            .----------------.
                            |                |
                            .----------------.
   .----------------.       |                |
   |                |       .----------------.
   .----------------.       |                |
   | Dev_A context  |------>'----------------'
   '----------------'          pasid table
   |                |          for Dev_A
   '----------------'
   | Dev_B context  |-->
   '----------------'  |    .----------------.
   |                |  |    |                |      .------------------.
   .----------------.  |    .----------------.      |                  |
   |                |  |    |                |      |                  |
   '----------------'  |    .----------------.      | Paging structure |
       Intel iommu     |    |                |      |  for Process B   |
      context table    |    .----------------.      |                  |
                       |    |     PASID Y    |----->'------------------'
                       |    .----------------.
                       |    |                |
                       |    .----------------.
                       |    |                |
                       |    .----------------.
                       v--->|                |
                            '----------------'
                               pasid table
                               for Dev_B

With this, dev_A has no means to access memory of process B and vice versa.

Best regards,
Lu Baolu

author
#13 lore raw

Hi,

Quoted 20 lines · Kevin
Quoted 20 lines

From: Lu Baolu [mailto:baolu.lu-VuQAYsv1563Yd54FQh9/[email protected]]
Sent: Wednesday, May 16, 2018 4:01 PM

Hi Joerg,

Thank you for looking at my patches.

Quoted 12 lines · Joerg Roedel
Quoted 5 lines · Lu Baolu

PATCH 4~9 implement per domain PASID table. Current per IOMMU PASID table implementation is insecure in the cases where multiple devices under one single IOMMU unit support PASID feature. With per domain PASID table, we can achieve finer protection and isolation granularity.

Hold on, we hat discussions in the past about doing a system-wide pasid space, so that every mm_struct with devices attached gets the same pasid across all devices it is talking to. Reason was that some devices (will) require this to work correctly. This goes into the opposite direction, so I am a bit confused here. Please explain, is this not longer necessary?

Quoted 8 lines
Quoted 3 lines

You are right. System-wide pasid space is necessary, hence PATCH 1~3 implement it. But PATCH 4~9 don't go into the opposite direction, it's designed to address another potential issue.

one thing you may want to highlight is, even with PATCH 4-9 it's still doing system-wide PASID space allocation. Just PASID table itself is kept per-device for isolation purpose as you described below, i.e. each device can access only those PASIDs which are allocated to itself while the allocation happens system-wide...

Yes, exactly.

Best regards,
Lu Baolu

Quoted 88 lines
Quoted 88 lines

With system-wide pasid space, we can use a system-wide pasid table, or just keep what we have now(per iommu unit pasid table). Both system-wide and per iommu unitpasid table mean that two devices might share a single pasid table. That will result in an issue.

For an example, device A is assigned to access the memory space of process A, and device B is assigned to access the memory space of process B. The dma remapping infrastructure looks like:

                                                     .------------------.
                             .----------------.      |                  |
                             |                |      |                  |
                             .----------------.      | Paging structure |
                             |     PASID X    |--|   |  for Process A   |
                             .----------------.  |   |                  |
                             |                |  --->'------------------'
     .----------------.      .----------------.
     |                |      |     PASID Y    |--|
     .----------------.      .----------------.  |
     | Dev_A context  |---|  |                |  |   .------------------.
     '----------------'   |  .----------------.  |   |                  |
     |                |   |  |                |  |   |                  |
     '----------------'   |  .----------------.  |   | Paging structure |
     | Dev_B context  |   -->|                |  |   |  for Process B   |
     '----------------'----->'----------------'  |   |                  |
     |                |         system-wide      v-->'------------------'
     .----------------.         pasid table
     |                |
     '----------------'
        Intel iommu
       context table

Since dev_A and dev_B share a pasid table, the side effect is that a flawed dev_A might access the memory space of process B (with pasid y). Vice versa,
a flawed dev_B might access memory space of process A (with pasid x).

What PATCH 4~9 do is to remove such possibility by assigning a pasid table for each pci device. Hence, the remapping infrastructure looks like:

                                                    .------------------.
                                                    |                  |
                            .----------------.      |                  |
                            |                |      | Paging structure |
                            .----------------.      |  for Process A   |
                            |     PASID X    |      |                  |
                            .----------------.----->'------------------'
                            |                |
                            .----------------.
                            |                |
                            .----------------.
                            |                |
                            .----------------.
   .----------------.       |                |
   |                |       .----------------.
   .----------------.       |                |
   | Dev_A context  |------>'----------------'
   '----------------'          pasid table
   |                |          for Dev_A
   '----------------'
   | Dev_B context  |-->
   '----------------'  |    .----------------.
   |                |  |    |                |      .------------------.
   .----------------.  |    .----------------.      |                  |
   |                |  |    |                |      |                  |
   '----------------'  |    .----------------.      | Paging structure |
       Intel iommu     |    |                |      |  for Process B   |
      context table    |    .----------------.      |                  |
                       |    |     PASID Y    |----->'------------------'
                       |    .----------------.
                       |    |                |
                       |    .----------------.
                       |    |                |
                       |    .----------------.
                       v--->|                |
                            '----------------'
                               pasid table
                               for Dev_B

With this, dev_A has no means to access memory of process B and vice versa.

Best regards,
Lu Baolu

maintainer
#14 lore raw
In reply to Lu Baolu
Quoted 3 lines · Lu Baolu

You are right. System-wide pasid space is necessary, hence PATCH 1~3 implement it. But PATCH 4~9 don't go into the opposite direction, it's designed to address another potential issue.

Okay, thanks for the clarification. Please send me a rebased version of this after the merge window I will review again.

Thanks,

Joerg

author
#15 lore raw

Hi,

Quoted 6 lines · Joerg Roedel
Quoted 3 lines · Lu Baolu

You are right. System-wide pasid space is necessary, hence PATCH 1~3 implement it. But PATCH 4~9 don't go into the opposite direction, it's designed to address another potential issue.

Okay, thanks for the clarification. Please send me a rebased version of this after the merge window I will review again.

Sure.

Best regards,
Lu Baolu